Loading...
Please wait, while we are loading the content...
Similar Documents
3-d Brain Image Registration Using Morphological Processing and Iterative Principal Axis Transform
| Content Provider | Semantic Scholar |
|---|---|
| Author | Results, Iii Conclusion, Iv Giardina, C. R. Daugherty, Erin Lončarić, Sven Dhawan, Atam P. |
| Abstract | two slices which may have occurred as a result of erroneous segmentation. A subsequent step of 3-D morphological opening is performed for the same reason. The described procedure segments out brain volume correctly in 90 % of MR slices and 97 % of PET slices while user correction is required for erroneously segmented slices. PET segmentation is more accurate because the skull is not present in PET scans. A shape-based interpolation is used to create a full volume with slice thickness equal to the slice pixel size. Typically, pixel size is of the order of 0.5 mm while the original slice thickness is of the order of 1 cm. Shape-distance maps are computed and cubic interpolation is used to add new slices at required positions. Interpolated grayscale image is then thresholded to get the full MR and PET binary volumes. The iterative principal axis algorithm is used to correlate MR brain volume and partial PET brain volume. In this work, an improved IPAR algorithm was used. The improved version uses the original idea [5] but has an improved iteration algorithm. To eliminate of problem of possible oscillating and not converging to a solution a weighting of partial solutions is introduced which insures convergence towards the solution. Assume that vector S i denotes the PAR parameters of the MR volume in the iteration i of the algorithm. The original algorithm used this vector to perform translations and rotations in order to align the volume. The improved iteration algorithm takes the previous value of the principal axis parameters in the account in order to prevent possible divergence from the solution. The actual PAR parameters at iteration i are computed as S 0 i = S i + (1 0)S i01 where 2 (0; 1). The result is more uniform and monotone convergence to the solution. A set of automatically extracted MR slices is shown in Figure 1. An example of matched MR and PET slices (original MR slice 10 and matching interpolated and transformed PET slice) is shown in Figure 2. For each slice a color image is produced for better presentation to user. The produced color images have the intensity of MR images and color determined by PET images. A new improved version of the IPAR procedure for registering 3-D MR and PET brain images has been presented. The improved registration procedure uses K-Figure 1: Extracted MR brain volume Figure 2: … |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://helga.zesoi.fer.hr/papers/embs93.ps.gz |
| Language | English |
| Access Restriction | Open |
| Subject Keyword | Algorithm Align (company) Apache Axis Axis vertebra CT scan Color image Convergence (action) Crystal structure Cubic Hermite spline Extraction Grayscale Color Map Image Slice Thickness Image registration Interpolation Imputation Technique Iteration Language Translations MATCHING Opening (morphology) Oxygen 100 % Gas for Inhalation PET/CT scan Pixel Polyethylene terephthalate Projection-slice theorem Scanning Segmentation action Solutions Thickness (graph theory) Web Slice monotone registration - ActClass |
| Content Type | Text |
| Resource Type | Article |
